Laboratory-based hyperspectral visible near-infrared reflectance spectral dataset of soil samples across a range of surface orientations

Duro, Alyssa1 ; Hirmas, Daniel2 ; Ajami, Hoori1 ; Billings, Sharon3 ; Li, Li4 ; Flores, Alejandro5 ; Guilinger, James6 ; Oleghe, Ewan7 ; Gimenez, Daniel7 ; Gray, Andrew1 ; Sullivan, Pamela8

Research facility: University of California, Riverside

Published Mar 19, 2024 on Dryad. https://doi.org/10.6086/D15091

Abstract

A custom-designed, 3-D printed sample array was used to present 681 homogenized soil samples packed into sample wells to a laboratory-based visible near-infrared (VNIR) hyperspectral imaging (HSI) reflectance spectrometer in a total of 91 different configurations of slope and aspect. Hyperspectral imaging was performed with a high-sensitivity sCMOS VNIR hyperspectral camera (MSV 500, Middleton Spectral Vision, Middleton, WI). Raw reflectance data were collected with FastFrame data acquisition software (Middleton Spectral Vision, Middleton, WI). After data were collected using FastFrame, data processing and analysis were performed in R using the R Scripts found on GitHub at github.com/aduro005/HSITopographicCorrectionRScripts. The design for this sample array can be found on GitHub at github.com/aduro005/HSITopographicCorrectionSampleWellArray.

README: Laboratory-based hyperspectral visible near-infrared reflectance spectral dataset of soil samples across a range of surface orientations

This README file was generated on 03/02/2024 by Alyssa Duro

This README file describes the R scripts available on GitHub at https://github.com/aduro005/HSITopographicCorrectionRscripts and the .RData files (available on Dryad Data Repository UCR at https://doi.org/10.6086/D15091. These R scripts and .RData files are associated with the HSI topographic correction method described in the article titled Topographic correction of visible near-infrared reflectance spectra for horizon-scale soil organic carbon mapping available at https://doi.org/10.1002/saj2.20612.

# ----------

# Some notes on this dataset:

This project began with the intention of calibrating several empirical regression equations to predict soil chemical properties from visible near-infrared (VNIR) reflectance spectra of soil surfaces positioned at different angles relative to a VNIR hyperspectral imaging (HSI) reflectance spectrometer. This is why the prefix “HSICalLib'' is used in file names and why this project is referred to in some places as the “HSI Calibration Library”. Later, the focus of the project narrowed, and the goal became developing a method for removing the influence of surface orientation from VNIR reflectance spectra. As a result, this project is also sometimes referred to as the “HSI Topographic Correction”.

The FastFrame software used to operate the HSI camera and scan stage outputs 2 data files after each scan. Together, these two files are sometimes referred to as the “raw data” for each scan. These two files have the same name, but one has the extension .hdr and the other is .raw. The file name and output location are input into the FastFrame software before a scan is performed. The .hdr file can be opened using Notepad. The .raw file is a 3-dimensional matrix with dimensions 471 (number of wavebands) x number of pixels in the lateral spatial dimension (also called “columns” and “samples”) x number of pixels in the direction of the scan stage movement (also called “rows” and “lines”).

\
The raw (unprocessed) data for this project is not included in this dataset, but the authors are happy to share it upon request. This raw dataset consists of 3,125 hyperspectral images (2 files per image) and is about 2 TB in size.

The most raw version of the soil VNIR reflectance spectra (observed, uncorrected) obtained at each slope, aspect configuration along with selected soil properties is called HSICalLib_b1_b30_p.RData and can be found on the UCR Dryad Data Repository. The soil reflectance spectra included here are averaged across all the pixels within each sample well. This file is output from the Step 7 of HSI Data Processing R script called HSICalLib_4_intmean_to_masterintmean_to_p.R (which can also be found on GitHub). Starting with this file, you can follow Steps 8-14 of HSI Data Processing (also on GitHub) to obtain all of the input files you need for ALL of the HSI Data Analysis workflow (Steps 1-4 and Final Plots) which was used for the HSI Topographic Correction. Even so, all of the files used during the HSI Topographic Correction that are output after HSICalLib_b1_b30_p.RData in the Data Processing and HSI Data Analysis workflows are included in the UCR Dryad Data Repository dataset. So, you can open any R script AFTER Step 7 of HSI Data Processing (also called HSICalLib_4_intmean_to_masterintmean_to_p.R), and the necessary input files are included on the UCR Dryad Data Repository dataset.

# ----------

# File folder structure:

Data Analysis

> Data

> Output Files

> Plots

> R scripts

# ---------------------------------------------------------------------------

# Description of the data

# ---------------------------------------------------------------------------

Hyperspectral imaging of 1178 homogenized soil samples was performed with a high-sensitivity sCMOS VNIR hyperspectral camera (MSV 500, Middleton Spectral Vision, Middleton, WI) in the Department of Environmental Sciences at University of California, Riverside.

Each soil sample was packed into sample wells, positioned under the hyperspectral camera, and imaged at 98 orientations (i.e., 7 slope x 14 aspect angles) using a custom designed and 3-D printed sample array. The sample array was designed such that sample wells could be presented to the spectrometer at 7 slope angles (0°, 10°, 20°, 30°, 40°, 50°, 60°). The design for this sample array is available on GitHub at https://github.com/aduro005/HSITopographicCorrectionSampleWellArray.

The sample array was aligned to an arbitrarily defined 0° N aspect and rotated at 15° intervals from 0° N to 90° E, and 180° S to 225° W using a protractor affixed to the scan stage under the spectrometer. The 195° to 270° W aspect values were converted to 165° to 90° E aspects during data processing (see HSICalLib_6a_aspect_correction.R). This was done so the aspect angle values only took on values from 0° to 180° in the development of the topographic correction method.

574 soil samples were provided by the NEON Initial Characterization Soils Archive at the University of Michigan Biological Station Sample Archive Facility in Ehlers (UMBS-SAFE) (https://mfield.umich.edu/soil_archive_request) and accompanying soil properties data were obtained from the NEON Data Portal (https://data.neonscience.org/).

450 soil samples were collected from Duke Farms in Hillsborough Township, New Jersey (https://www.dukefarms.org/) and accompanying soil properties data were provided by the Department of Environmental Sciences at Rutgers University.

57 soil samples were collected from locations in the Santa Ana Mountains, California that were affected by wildfire and accompanying soil properties data was provided by the Department of Environmental Sciences at University of California, Riverside (http://www.thegraylab.org/).

97 soil samples were a laboratory standard soil from the Pedology Laboratory in the Department of Environmental Sciences at University of California, Riverside.

# ----------

# List of files in the Output Data folder:

HSICalLib_wavevec.RData

HSICalLib_b1_b30_p.RData

HSICalLib_b1-b30_20230223_p_gold.RData

HSICalLib_20230223_s0-s60_cosIL_all.RData

HSICalLib_20230223_b1-b30_pga.RData

HSICalLib_20230223_b1-b30_pga_melt.RData

HSICalLib_20230223_b1-b30_pga_refI_melt.RData

HSICalLib_20230223_b1-b30_pga_dI_melt.RData

HSICalLib_20230310_rutgerssamples_rand1.RData

HSICalLib_20230613_globaldI_predict_lm_pga_dI_refI_melt_slm_dIp_dIc_2.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_1.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_w_a_2.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_w_s_2.RData

HSICalLib_20230613_pga_slm_cosIL.RData

HSICalLib_20230613_spectralstats_coscor.RData

HSICalLib_20230613_spectralstats_ccor.RData

HSICalLib_20230613_globaldI_OCpredict_refI_p.RData

HSICalLib_20230613_globaldI_OCpredict_obsI_p.RData

HSICalLib_20230613_globaldI_OCpredict_dIc_p.RData

HSICalLib_20230613_globaldI_PLSR_log10volC_sstatdf_tr_ref.RData

HSICalLib_20230613_globaldI_PLSR_log10volC_sstatdf_tr_dIc.RData

HSICalLib_20230613_globaldI_PLSR_log10volC_sstatdf_dIc.RData

HSICalLib_20230613_globaldI_PLSR_log10volC_sstatdf_obs.RData

HSICalLib_20230613_globaldI_PLSR_log10volC_summarystats_OC.RData

# ----------

# List of abbreviations:

obsI = observed, uncorrected reflectance intensity

refI = reference reflectance intensity, mean of all reflectance spectra observed for each soil sample across all aspects at zero slope, represents a soil sample’s expected reflectance spectrum when the effect of surface orientation on reflectance is absent

dI = delta intensity, obsI - refI

dIp = predicted dI, this value is predicted by a multiple linear regression model trained to predict dI from slope, aspect, wavelength, and their interaction terms

dIc = dI-corrected VNIR reflectance intensities, obsI - dI

coscorI = cosine corrected reflectance intensities

ccorI = C corrected reflectance intensities

# ----------

# Specific information for data file:

HSICalLib_wavevec.RData

# Name and type of R object: wavevec (vector, numeric)

# Number of observations (rows): 471

# Description of observations:

Each observation is a wavelength (λ) in units of (nm). Wavelengths corresponds to reflectance intensity observations (measurements) made at each waveband for reflectance spectra collected using the high-sensitivity sCMOS VNIR hyperspectral camera (MSV 500, Middleton Spectral Vision, Middleton, WI) in the Department of Environmental Sciences at University of California, Riverside. This spectrometer measures 471 reflectance intensities (i.e., reflectance intensities are measured at 471 wavebands) between 400 and 1000 nm wavelengths. The 471 observations in this vector (“wavevec”) are the wavelengths corresponding to each waveband.

# Missing data values (NA): None

# Number of variables (columns): 1

# Description of variables:

wavelength (nm)

# Related data files:

Any .hdr file resulting from a scan with this spectrometer contains the information contained in HSICalLib_wavevec.RData. However, the .hdr files also contain other information, so this vector was made in R by Alyssa Duro so the wavelength values corresponding to each waveband were accessible on their own in a .RData file.

# R script that outputs this file:

HSICalLib_3_intensities_to_intmean_intsd.R

# ----------

# Specific information for data file:

HSICalLib_b1_b30_p.RData

# Name and type of R object: p (data frame)

# Number of observations (rows): 115,444 reflectance spectra

# Description of observations:

Mean reflectance spectra for each soil sample at each orientation along with selected soil properties data and sample identifiers. This is the most raw form of the data included in this dataset. 1178 soil samples * 98 configurations = 115,444

# Missing data values (NA): Some soil properties data are not available for all soil samples resulting in NA’s.

# Number of variables (columns): 486

# Description of variables:

obsI*[wavelength]* (numeric): observed (uncorrected) reflectance intensities measured at 471 wavebands, together, these 471 values represent the average reflectance spectrum for a single soil sample at a single orientation

slope (integer, degrees): angle between the scan stage and the soil surface

aspect (integer, degrees): angle clockwise from N

batch (integer, 1-30): soil samples were imaged in groups of 40 at a time

well (integer, 1-40): indexed location of the soil sample in the sample well array

HSInumber (integer, 1-1141): unique soil sample identifier

HSIPackedDensity (numeric, g/cm3): mass soil sample per volume sample well

sandTotal (numeric, %): sand (only available for samples from the NEON archive)

siltTotal (numeric, %): silt (only available for samples from the NEON archive)

clayTotal (numeric, %): clay (only available for samples from the NEON archive)

OC (numeric, %): soil organic carbon (by weight)

archive (character): source of the soil sample and soil properties data

adod (numeric, unitless): air dried soil mass / oven dried soil mass

volC (numeric, %): soil organic carbon (by volume)

log10volC (numeric): log10(volC)

batchwellID (character): unique reflectance spectra identifier

# R script that outputs this file:

HSICalLib_4_intmean_to_masterintmean_to_p.R

# ----------

# Specific information for data file:

HSICalLib_b1-b30_20230223_p_gold.RData

# Name and type of R object: p_gold (data frame)

# Number of observations (rows): 107,486 reflectance spectra

# Description of observations:

Same as HSICalLib_b1_b30_p.RData (output from R script 4) except the reflectance spectra (rows) with unusually large or small obsI OR dI values have been identified and removed as imaging errors.

# Missing data values (NA): Missing values occur when soil properties data are not available for some soil samples. There is spectral data for all soil samples, but some soil properties were not measured for all soil samples.

# Number of variables (columns): 486

# Description of variables:

Same as HSICalLib_b1_b30_p.RData (output from R script 4)

# Related data files:

HSICalLib_b1_b30_p.RData

# R script that outputs this file:

HSICalLib_5b_p_dI_cleaning.R

# ----------

# Specific information for data file:

HSICalLib_20230223_s0-s60_cosIL_all.RData

# Name and type of R object: cosIL (numeric, data frame)

# Number of observations (rows): 91 orientations (7 slopes * 13 aspects)

# Description of observations:

Each row contains the constants needed to perform the cosine correction and C correction for 1 of 91 possible combinations of slope and aspect.

# Missing data values (NA): None

# Number of variables (columns): 16

# Description of variables:

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

z1: zenith angle (degrees) between light bank 1 and the HSI camera, varies with slope, light bank 1 = N = 0 azimuth

z2: zenith angle (degrees) between light bank 2 (S) and the HSI camera, varies with slope, light bank 2 = S = 180 azimuth

meanz: average of z1 and z2 (this is the one used for the paper)

cosz1: cosine of z1

cosz2: cosine of z2

cosmeanz: cosine of (meanz)

meancosz: average of cos(z1) and cos(z2)

cosIL1: cos( illumination angle light bank 1 (IL1) ) = cos(z1)*cos(slope) + sin(z1)*sin(slope)*cos(azimuth-aspect)

cosIL2: cos( illumination angle light bank 2 (IL2) ) = cos(z2)*cos(slope) + sin(z2)*sin(slope)*cos(azimuth-aspect)

meancosIL: average of cosIL1 and cosIL2

r1: cos(z1) / cosIL1

r2: cos(z2) / cosIL2

rmeans: ( cos(meanz) ) / (meancosIL)

rcosmeans: (meancosz) / (meancosIL) (this is the one used for the paper)

# Related data files:

HSICalLib_20230613_pga_slm_cosIL.RData

# R script that outputs this file:

HSICalLib_6b_cosIL_calculation.R

# ----------

# Specific information for data file:

HSICalLib_20230223_b1-b30_pga.RData

# Name and type of R object: pga (data frame)

# Number of observations (rows): 99,537 reflectance spectra

# Description of observations:

Same as HSICalLib_b1_b30_p.RData (output from R script 4) except the number of spectra was reduced during the “aspect correction” (HSICalLib_6a_aspect_correction.R)

# Missing data values (NA): None

# Number of variables (columns): 486

# Description of variables:

Same as HSICalLib_b1_b30_p.RData

# Related data files:

HSICalLib_b1_b30_p.RData

# R script that outputs this file:

HSICalLib_7a_observedI

# ----------

# Specific information for data file:

HSICalLib_20230223_b1-b30_pga_melt.RData

# Name and type of R object: pga_melt (data frame)

# Number of observations (rows): 46,881,927 reflectance intensities

# Description of observations:

“Long” version of HSICalLib_20230223_b1-b30_pga.RData where wavelength is a variable. 99,537 reflectance spectra (from HSICalLib_20230223_b1-b30_pga.RData) * 471 wavebands = 46,881,927 reflectance intensities

# Missing data values (NA): None

# Number of variables (columns): 5

# Description of variables:

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

HSInumber: see HSICalLib_b1_b30_p.RData

wavelength: see HSICalLib_wavevec.RData

obsI: Same values reported in HSICalLib_b1_b30_p.RData except now wavelength is a variable. These values have not been “corrected”.

# Related data files:

HSICalLib_20230223_b1-b30_pga.RData

# R script that outputs this file:

HSICalLib_7a_observedI

# ----------

# Specific information for data file:

HSICalLib_20230223_b1-b30_pga_refI_melt.RData

# Name and type of R object: pga_refI_melt (data frame)

# Number of observations (rows): 46,881,927 reflectance intensities

# Description of observations:

Same as HSICalLib_20230223_b1-b30_pga_melt.RData except reference reflectance intensities (refI) are reported instead of obsI.

# Missing data values (NA): None

# Number of variables (columns): 5

# Description of variables:

Same as

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

HSInumber: see HSICalLib_b1_b30_p.RData

wavelength: see HSICalLib_wavevec.RData

refI: These values represent the average reflectance spectrum measured for each soil sample across all aspect positions at zero slope. There is only 1 reference spectrum per soil sample.

# Related data files:

HSICalLib_20230223_b1-b30_pga_melt.RData

# R script that outputs this file:

HSICalLib_7b_referenceI

# ----------

# Specific information for data file:

HSICalLib_20230223_b1-b30_pga_dI_melt.RData

# Name and type of R object: pga_dI_melt (data frame)

# Number of observations (rows): 46,881,927 reflectance intensities

# Description of observations:

Same as HSICalLib_20230223_b1-b30_pga_melt.RData except dI values are reported instead of obsI

# Missing data values (NA): None

# Number of variables (columns): 5

# Description of variables:

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

HSInumber: see HSICalLib_b1_b30_p.RData

wavelength: see HSICalLib_wavevec.RData

dI: actual (measured) delta (“change in”) reflectance intensity = obsI - refI

# Related data files:

HSICalLib_20230223_b1-b30_pga_melt.RData

HSICalLib_20230223_b1-b30_pga_refI_melt.RData

# R script that outputs this file:

HSICalLib_7c_dI_calculation

# ----------

# Specific information for data file:

HSICalLib_20230310_rutgerssamples_rand1.RData

# Name and type of R object: rand1 (numeric, vector)

# Number of observations (rows): 50

# Description of observations:

A randomly chosen subset of 50 soil samples (out of the 450 samples collected from Duke Farms and imaged using HSI) were included in the topographic correction study due to these soil sample properties all being very similar while making up a large portion of the training data. This vector contains the HSI numbers for these 50 randomly chosen soil samples (all from the Rutgers archive).

# Missing data values (NA): None

# Number of variables (columns): 1

# Description of variables:

HSInumber: see HSICalLib_b1_b30_p.RData

# Related data files:

HSICalLib_20230613_globaldI_predict_lm_pga_dI_refI_melt_slm_dIp_dIc_2.RData

# R script that outputs this file:

HSICalLib_8b_dI_predict_global_final.R

# ----------

# Specific information for data file:

HSICalLib_20230613_globaldI_predict_lm_pga_dI_refI_melt_slm_dIp_dIc_2.RData

# Name and type of R object: pga_dI_refI_melt_slm_dIp_dIc_2 (data frame)

# Number of observations (rows): 22,678,179

# Description of observations:

Each row is a reflectance intensity for a single soil sample at a single orientation at a single wavelength. Same as HSICalLib_20230223_b1-b30_pga_melt.RData except the number of observations was reduced by selecting ONLY the HSInumbers (spectra) for the 681 soil samples included in this study. This data frame is output AFTER training and evaluating the dI+ correction wherein a multiple linear regression model was trained to predict dI using slope, aspect, wavelength, and their interactions as predictor variables. This model was evaluated to get predicted dI (dIp), then dIp was used to adjust (aka “correct”) obsI resulting in dI-corrected intensities (dIc). If the model was a perfect predictor, then dIp would equal dI AND dIc would equal refI.

# Missing data values (NA): None

# Number of variables (columns): 9

# Description of variables:

Same as HSICalLib_20230223_b1-b30_pga_melt.RData except dI, refI, dIp, and dIc columns have been added.

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

HSInumber: see HSICalLib_b1_b30_p.RData

wavelength: see HSICalLib_wavevec.RData

obsI: see HSICalLib_20230223_b1-b30_pga_melt.RData

dI: see HSICalLib_20230223_b1-b30_pga_dI_melt.RData

refI: see HSICalLib_20230223_b1-b30_pga_refI_melt.RData

dIp: predicted dI resulting from evaluation of the dI+ multiple linear regression model, if the model was a perfect predictor, dIp would equal dI

dIc: dI-corrected reflectance intensities = obsI - dIp, if the dI correction was successful, dIc should equal refI

# Related data files:

HSICalLib_20230223_b1-b30_pga_melt.RData

# R script that outputs this file:

HSICalLib_8b_dI_predict_global_final.R

# ----------

# Specific information for data file:

HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

# Name and type of R object: spectralstats_dIc_2 (data frame)

# Number of observations (rows): 48,149

# Description of observations:

Number of spectra compared when reflectance intensities are grouped this way before calculating objective functions. RMSE, NSE, and KGE are calculated for every soil sample at every configuration across all wavelengths. These dIc values were obtained by evaluating the dI+ multiple linear regression model (i.e., the one that includes slope, aspect, wavelength, and all their interactions as predictor variables).

# Missing data values (NA): None

# Number of variables (columns): 15

# Description of variables:

slope: see HSICalLib_b1_b30_p.RData

aspect: see HSICalLib_b1_b30_p.RData

HSInumber: see HSICalLib_b1_b30_p.RData

RMSE_obs: Root mean squared error (obsI vs refI) tells how far obsI is from refI, RMSE=0 suggests surface orientation has no effect

RMSE_dIc: Root mean squared error (dIc vs refI) tells how far dIc is from refI, RMSE=0 suggests dI correction was successful

NSE_obs: Nash-Sutcliffe efficiency (obsI vs refI) tells if obsI is closer to mean refI or refI, NSE=1 suggests surface orientation has no effect

NSE_dIc: Nash-Sutcliffe efficiency (dIc vs refI) tells if dIc is closer to mean refI or refI, NSE=1 suggests dI correction was successful

KGE_obs: Kling-Gupta efficiency (obsI vs refI), same as NSE

KGE_obs_r: Pearson correlation coefficient (obsI vs refI), component of KGE

KGE_obs_beta: mean obsI / mean refI (obsI vs refI), component of KGE, ratio of the means of obsI and refI, beta=1 suggests surface orientation has no effect

KGE_obs_alpha: standard deviation(obsI) / standard deviation(refI), component of KGE, ratio of the standard deviations of obsI and refI, alpha=1 suggests surface orientation has no effect

KGE_dIc: Kling-Gupta efficiency (dIc vs refI), same as NSE

KGE_dIc_r: Pearson correlation coefficient (dIc vs refI), component of KGE

KGE_dIc_beta: mean dIc / mean refI (dIc vs refI), component of KGE, ratio of the means of dIc and refI, beta=1 suggests dI correction was successful

KGE_dIc_alpha: standard deviation(dIc) / standard deviation(refI), component of KGE, ratio of the standard deviations of dIc and refI, alpha=1 suggests dI correction was successful

# Related data files:

HSICalLib_20230613_globaldI_spectralstats_dIc_w_s_2.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_w_a_2.RData

HSICalLib_20230613_spectralstats_coscor.RData

HSICalLib_20230613_spectralstats_ccor.RData

# R script that outputs this file:

HSICalLib_8b_dI_predict_global_final.R

# ----------

# Specific information for data file:

HSICalLib_20230613_globaldI_spectralstats_dIc_1.RData

# Name and type of R object: spectralstats_dIc_1 (numeric, data frame)

# Number of observations (rows): 48,149

# Description of observations:

Number of spectra compared when reflectance intensities are grouped this way before calculating objective functions. RMSE, NSE, and KGE are calculated for every soil sample at every configuration across all wavelengths. These dIc values were obtained by evaluating the dI multiple linear regression model (i.e., the one that includes ONLY slope, aspect, and wavelength as predictor variables).

# Missing data values (NA): None

# Number of variables (columns): 15

# Description of variables:

Same as HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

# Related data files:

HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

# R script that outputs this file:

HSICalLib_8b_dI_predict_global_final.R

# ----------

# Specific information for data file:

HSICalLib_20230613_globaldI_spectralstats_dIc_w_s_2.RData

# Name and type of R object: spectralstats_dIc_w_s_2 (data frame)

# Number of observations (rows): 2,826

# Description of observations:

Number of sample groups compared when reflectance intensities are grouped this way before calculating objective functions. RMSE, NSE, and KGE are calculated across all aspects and soil samples at each wavelength, slope combination (471 wavelengths * 6 slopes = 2,826 results for each objective function). Same as HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData BUT reflectance intensities were grouped in a different way before calculating objective functions.

# Missing data values (NA): None

# Number of variables (columns): 14

# Description of variables:

slope: see HSICalLib_b1_b30_p.RData

wavelength: see HSICalLib_wavevec.RData

RMSE_obs: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

RMSE_dIc: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

NSE_obs: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

NSE_dIc: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_obs: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_obs_r: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_obs_beta: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_obs_alpha: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_dIc: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_dIc_r: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_dIc_beta: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

KGE_dIc_alpha: see HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

# Related data files:

HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData

HSICalLib_20230613_globaldI_spectralstats_dIc_w_a_2.RData

# R script that outputs this file:

HSICalLib_8b_dI_predict_global_final.R

# ----------

# Specific information for data file:

HSICalLib_20230613_globaldI_spectralstats_dIc_w_a_2.RData

# Name and type of R object: spectralstats_dIc_w_a_2 (data frame)

# Number of observations (rows): 6,123

# Description of observations:

Number of sample groups compared when reflectance intensities are grouped this way before calculating objective functions. RMSE, NSE, and KGE are calculated across all slopes and soil samples at each wavelength, aspect combination (471 wavelengths * 13 aspects = 6,123 results for each objective function). Same as HSICalLib_20230613_globaldI_spectralstats_dIc_2.RData BUT reflectance intensities were grouped in a different way before calculating objective functions.